Slator Co-Founder and Commercial Director Andrew Smart opened SlatorCon Remote June 2023 by making the case for meeting online. Quick, informative, cost-effective, and with more than 400 registrants from over 40 countries, it also provides global reach.
Notably, “client-side participation has continued to climb and is now at 54%, surpassing language service providers (LSPs),” Smart said, with attendees representing enterprises from across all industries.
With more moderate inflation and interest rates, plus a surge of feature launches incorporating AI and large language models (LLMs), the language industry has, in Smart’s opinion, a solid case for optimism.
“I hope you’re not suffering from complete LLM/AI fatigue yet,” Slator Co-Founder and Managing Director Florian Faes joked as he began a rundown of developments, opportunities, and risks related to ChatGPT and translation.
More specifically, Faes said, the industry is now seeing the rise of general purpose language models, as opposed to specific purpose language models built specifically for machine translation (MT).
OpenAI’s November 2022 launch of ChatGPT could be considered the innovation trigger of the current hype cycle, which Faes suggested might last through 2027. With their many use cases, including multilingual text generation, MT quality estimation, and text editing — among others — general purpose language models will simply have a broader impact than the launch of neural MT in 2016.
In Q2 2023 alone, open-source machine learning model repository HuggingFace launched Hugging Chat, with MT; Google made Bard available in the US and Australia; and Meta challenged OpenAI’s speech-to-text model Whisper.
According to Slator’s 2023 Language Industry Market Report survey of LSP leaders, integrating LLMs into production is not yet “systemic or transformational” — 60% of respondents had not tried it. But the potential is there, and time is of the essence.
“I don’t think you can afford not to start using it and start building some expertise in it,” Faes said, adding that market cautiousness is fair but probably overdone. “The language industry has proven time and time and again that it’s very adaptive and resourceful and creative in finding new ways to find solutions for clients.”
In-Person Interpreting and Other Human Activities
More than 50% of Propio Language Services’ clients are healthcare providers, and so far, CEO and Keynote Speaker Marco Assis said, very few clients have asked about ChatGPT or LLMs, despite the buzz.
“When you work in a regulatory framework, when you work with critical services, we’re talking about slower adopters for a good reason,” Assis explained. “We are not in a position where we can make a mistake.”
Years ago, Assis thought that in-person interpretation would be replaced by other, more technologically advanced modalities. But the shift in focus to user experience has actually led to in-person interpreting becoming one of the most resilient segments of the business.
AI’s impact on the business landscape prompted Dell Technologies to invest in its people. Dell’s Senior Manager of Globalization Program Management, Aki Hayashi, explained that the company asked for expert advice from three trusted vendors — Welocalize, Lionbridge, and RWS — when creating a new career development program for project managers.
Of course, Dell also brought in their own internal resources, RWS Global Program Manager Genevieve D’Acquisto noted. The program succeeded thanks in part to the fact that participants “had either worked with us on the vendor side or had existing long-term relationships with folks on the vendor side that really made them comfortable taking that leap.”
Over 80% of Fortune 500 companies work with the holistic business platform ServiceNow. Globalization Deployment Director Alex Coope’s internationalization group helps partners understand how to design solutions that work with other languages both technically and linguistically.
“Just because we add 22 languages out of the box, that doesn’t mean that they couldn’t add more,” Coope explained. “We’ve got customers who have tens of languages, maybe even 100 plus languages. It doesn’t matter. There’s no technical limitation as to why they shouldn’t be able to do that.”
Doubling Down on Dubbing
Traditionally associated with OTT streaming services and film, dubbing is gaining ground in educational content, where comprehension is the priority. Leading a group discussion on AI Dubbing with a Human Touch, Dubformer Co-Founder and CEO Anton Dvorkovich remarked, “We have never seen a lot of online content on air dubbed already. But we expect more than a million units of AI-dubbed media and e-learning content on air in the year 2024.” He credits the speed in adoption to developments in transformer neural networks and the global hype around LLMs.
Slator’s own Head of Research Anna Wyndham moderated a panel on the future of AI dubbing. Subhabrata Debnath, Co-Founder and CTO of India-based NeuralGarage, explained how his company breaks down spoken words into small segments (phonemes) and their corresponding lip movements to create a language-agnostic “dictionary.” Debnath said he thinks adoption will be rapid once viewers see a state-of-the-art solution, as it will become harder to ignore the issues with traditional lip sync.
Blanc Technologies Co-Founder Richard Pavlovskiy identified voice-cloning as a definite goal for his company and predicted that, within the next 10 years, “probably any type of video” will be “translated into all languages.”
For now, Blanc Technologies is focused on cost savings. “It’s extremely, extremely expensive,” Pavlovskiy said of the current method. “If we look at the market of full video translations, when you just upload the video, for example, to the service and get the translation back, it’s usually around a few hundred dollars per minute of video. We managed to get it to the range of USD 3-4 per minute.”
FIFA’s Goals and Sişecam’s Glass Ceiling
In a panel discussion with DeepL Customer Success VP Céline Daley, FIFA’s Wilma Ritter (Head of Language Center) and Marco Studer (PM, Translation Technologist, and German Translator) explored how the governing body of world football worked with AI-based translation platform DeepL to keep up with the “super fast-moving FIFA.”
FIFA decided to add three more languages (Arabic, Portuguese, and Russian) to its original four official languages (English, French, German, and Spanish). And FIFA’s decision to hold the 2026 World Cup in North America means the team must plan on improving localization efforts during the next 12-18 months.
In addition to localizing content into Spanish for Mexico and Canadian French, Ritter said, “the words ‘soccer’ and ‘football’ are already creating some challenges.”
From glassware to glass fiber, Turkey-headquartered glass manufacturer Sişecam sells its products in more than 150 countries — and managed to scale its sales network globally without a centralized localization strategy.
So when Sişecam’s first localization expert, Global Content and Localization Program Manager Olga Hergül, came on board in 2019, she decided to focus on localizing Turkish source content for the company’s 24,000 employees in 14 countries.
Coming full circle on the topic of LLMs in translation, Hergül shared her only observation from her experience thus far: “GPT transforms content in a very particular way … and this can lead to problems with style and the tone of voice,” she said. “That really highlighted the importance of having style guides in place.”
For those who missed SlatorCon Remote June 2023 in real-time, recordings will be available in due course via our Pro and Enterprise plans.